Materials and Methods .1 Study sites and plots

The original study design included 24 permanent study plots with 3 replicate plots at 8 elevational steps about every 500 m in elevational distance on the eastern Andean slope in Napo province, Ecuador (Salazar et al., submitted manuscript). This gradient spans localities from lowland forests in the vicinity of Río Napo (Reserva Jatun Sacha) at 500 m via Reserva Biósfera Sumaco, Reserva Ecológica Antisana, Estación Biológica Yanayacu, and Guango Lodge up to highest elevations at Reserva Ecológica Cayambe-Coca close to timberline at 4000 m (Unger et al., 2010; Salazar et al., submitted manuscript). Plots were located in natural slope forests, avoiding anthropogenically or naturally disturbed habitats as well as special microhabitats such as ravines or ridges. However, at 1500 m and 3000 m, single plots were atypical and these elevations were therefore excluded from the present study.

Study plots were of 20 m x 20 m (400 m²), a plot size that has previously been used for surveys of local tree (Homeier et al., 2010a) and fern diversity (e.g, Kessler 2001; Kluge et al., 2006) because it is large enough to be representative but also small enough to be ecologically homogeneous (Kessler and Bach, 1999). At those elevational levels where the abundance of terrestrial ferns was very high, the measurements of biomass and productivity of the most abundant species were restricted to a subplot of 10 m x 10 m embedded in the main plot and afterwards recalculated to the original plot size. This was the case in single plots at 1000 m, 2000 m, and 2500 m. Destructive sampling (herbarium specimens, leaves and rhizomes for biomass measurements) was conducted on fern individuals outside of the study plots. All species were collected for identification at the herbarium of the Pontificia Universidad Católica del Ecuador (QCA) and by specialists for difficult groups. We excluded epiphytic ferns from the study because the inaccessibility of the canopy implied that most individuals could be accessed, tagged, and remeasured.

Chapter 4

65

Plots at 500-2000 m were established in 2005-2008 for studies of trees (Homeier et al., 2010b; Unger et al., 2010, 2012) while plots at 2500-4000 m were established for the present study in 2009 and 2010. Field sampling campaigns took place in 2009 (May-August), 2010 (April-July), and 2011 (April-July). One plot each at 3500 m and 4000 m suffered human disturbance in the first study year and had to be relocated.

4.3.2 Diversity measurements

Species richness was expressed as the number of terrestrial fern species per 400 m² plot.

To test H2 we needed a richness measure that was independent of the number of fern individuals in a plot. We used Fisher’s Alpha, a scalar that accounts for differences in numbers of individuals and hence removes the sampling effect (Fisher et al., 1943;

Hubbell, 2001; Kaspari et al., 2003). We did not use individual-based rarefaction for this purpose because the lowest number of individuals in some plots (N = 11) was below that of the number of species in some other plots and would have led to the exclusion of 96% of the data points. Evenness of species abundances per plot was calculated using Simpsons (1949) evenness index E_1/D (Magurran, 2004; Tuomisto, 2012).

Source populations were defined as species records in a given plot that included at least one mature and fertile individual or, in a few species, a mature individual that showed vegetative reproduction, e.g. via buds (Grytnes et al., 2008; Kessler et al., 2011).

Accordingly, sink species were all those not able to reproduce in a given plot during the observation period.

4.3.3 Productivity measurements

Actual evapotranspiration (AET) was calculated using Turc’s formula (Turc 1954, González-Espinosa et al., 2004), where AET = P / [0.9 + (P/L)²]^1/2 with L = 300 + 25 T + 0.05 T³, P = mean annual precipitation, and T = mean annual temperature. Precipitation and temperature values were compiled from the WorldClim data bank (Hijmans et al., 2005), since there are no climate stations in the vicinity of the study sites.

Aboveground Biomass Increment of Trees (AGBItrees) was measured as stem increment measurements. In all plots, all stems with dbh ≥10 cm were equipped with dendrometer tapes (type D1; UMS, Munich, Germany) that were monitored at least once per year for stem diameter growth. Tree height was measured with a Vertex IV height meter and a T3 transponder (Haglöf, Langsele, Sweden) in April/ May 2011. AGBI_trees was calculated as

Chapter 4

66

the sum of all individual tree increments applying the allometric equation of Chave et al.

(2005) for tropical wet forests, with stem diameter, wood specific gravity (WSG) and tree height as parameters. WSG data for the tree species were obtained from Chave et al.

(2006) or, in cases of missing information on species, genera or family means of WSG were calculated from the same source and applied to the respective species. Growth monitoring started in 2005 in the plots at 500-2000m and data on ABGItrees of these plots were taken from Unger et al. (2012). ABGItrees of the plots at 2500-4000m was calculated over the period 2009-2012. A more detailed description of the ABGItrees calculation is given in Unger et al. (2012).

Aboveground Biomass Increment of Ferns (AGBI_ferns) was measured as the sum of annual leaf and rhizome biomass increment of all terrestrial fern individuals in each plot over two consecutive years (2009-2010, 2010-2011). At the beginning of the study, we marked the rhizomes of all fern individuals in the study plots with numbered plastic tags, measured rhizome length and frond length with a tape measure, and counted the number of fronds. Each petiole was marked with non-corrosive wire so that in the consecutive field phases turnover of leaves and productivity could be unambiguously assessed. In subsequent field campaigns, newly developed fern individuals and leaves were marked and measured in similar ways.

To relate leaf length to leaf biomass, three healthy mature fronds with few or no epiphylls were sampled for each species at each elevational level . Leaves were scanned and their leaf-area and length were measured from the digital image using WinFolia (Regent Instruments Inc. Quebec, Canada) and ImageJ (Rasband, 1997–2009). After leaves were dried at 65°C, their mean tissue density was evaluated by weighting (accuracy 0.1 g).

Individual level leaf length (x) and foliar tissue density (y) were related by averaging values of the fronds collected for all species at all elevations using the linear regression equation y = 0.0002*x^2.2854 (R²=0.87, p<0.001). Number of fronds and leaf-length were combined with the tissue density values to calculate total biomass and biomass increment, respectively.

A similar procedure was applied to rhizomes, with three specimens collected, weighted, and used to relate field measurements of rhizome length. For partly subterranean rhizomes,

Chapter 4

67

biomass increment was quantified by measuring the distance between the petioles of the old and newly formed leaves.

Overall, biomass values were obtained from 399 vouchers of 82 species. Nine rare species (1-3 individuals per plot) were only found within the plots and the values for these species were taken from species of the same genus that were the most comparable in size and texture.

Because productivity measures were unavailable for two plots for the first year of the study, we conducted all analyses with the data from the second year only. However, for the remaining 16 plots, productivity values were highly correlated between the two study years (Spearman’s Rank Correlation, r = 0.98, p < 0.001), showing that our data are representative for the respective plots.

4.3.4 Data analyses

Prior to analyses, AGBIferns was log-transformed to approach normality. To assess if this affected our results, we also conducted the analyses with untransformed data using non-parametric statistics and obtained qualitatively similar results (data not shown).

For analyses along the elevational transect, we used the raw values of species richness, productivity, etc. per plot.

For analyses within elevational belts, we calculated the mean value of each parameter from the three plots at each elevation and then expressed the values of each plot relative to that mean. In this way, we were able to combine plots from different elevations (which had different absolute values) by using a common relative measure, resulting in n=18 for the analyses. AET was excluded from this analysis because it was calculated as a standard value for each elevation based on climatic data and hence provided no information on variability at a given elevation.

To test H1, we used ordinary least-squares (OLS) regression with Species Richness as dependent variable and AET, AGBItrees, and log(AGBIferns) as explanatory variables.

To test H2, we used ordinary least-squares (OLS) regression with Fisher’s Alpha as dependent variable and AET, AGBItrees, and log(AGBIferns ) as explanatory variables. H2 is supported if there is no or a negative relationship of a productivity measure (using only those measures not rejected in H1) to fern diversity as measured by Fisher’s Alpha, a

Chapter 4

68

diversity index that accounts for differences in numbers of individuals (Kaspari et al., 2000, 2003; Beck et al., 2011).

To test H3, we used ordinary least-squares (OLS) regression with Fisher’s Alpha as dependent variable and Total Number of Fern Individuals as explanatory variable. Further, because this hypothesis predicts that productivity increases species richness via an increase in the number of individuals, we tested for a covariance between Number of Fern Individuals and log(AGBIferns) using Spearman’s Rank Correlation.

To test H4a, we used ordinary least-squares (OLS) regression with Evenness as dependent variable and log(AGBIferns) as explanatory variable. We also assessed whether Evenness was correlated to Fisher’s Alpha using Spearman’s Rank Correlation. To test H4b, we conducted ordinary least-squares (OLS) regression using Fisher’s Alpha as dependent variable and Mean Number of Individuals per species and plot as explanatory variable.

To test H5, we re-ran all analyses for H1-H4 after excluding sink species from each plot.

The hypothesis is supported if results of the analyses regarding H1-H4 are affected when sink populations are excluded.

Analyses were run with the statistical platform R (R Development Core Team 2012).